Should teachers worry about ChatGPT?

ChatGPT artificial intelligence chatbot can, among other things, generate articles and write computer code. Since it went public for testing late last year, it has raised concerns about students using ChatGPT to complete their homework and prompted some public high schools to ban it and college professors to alter their course assignments.

Ted Underwood is Professor of English and Information Science and Associate Dean for Academic Affairs in the College of Information Sciences. He recently commented in Inside Higher Ed on how he views the place of technology in higher education. He spoke with Judy Haikal, Arts and Humanities Editor at the News Desk.

What is ChatGPT and how is it different from previous versions of chatbots?

The discussion tends to focus on ChatGPT because this product was made widely available for free last fall – and it was a little easier to use than previous language models. But ChatGPT is far from unique.

ChatGPT is based on technology that has been around in one form or another since OpenAI released the first version of its pre-trained generative transformers in 2018. The basic idea is that the model is trained to predict the next word in an observable sequence of words. Then when you type a short syllable — a “prompt” — the model can predict the next word in the sequence, then the next, and so on. To do a really good job, the model needs to recognize high-level patterns and act as if it understands the language. Because models like this grew better upon generalization as the researchers increased the size of the model, they are sometimes called “large language models.” They are also called “generative AI” because they not only analyze texts but use what they have learned to generate new ones.

ChatGPT improves on previous versions of this technology by specifically training a model to treat and respond to prompts as turns in a conversation, rather than just continuing your statement.

ChatGPT is not unique in this; Similar models have been released by Google, Meta, and Anthropic. And OpenAI itself recently released GPT-4, which is better than ChatGPT.

Should teachers be concerned about students using ChatGPT or other AI writing software to write their research papers, or should they consider how AI applications can be used as teaching tools to help students learn?

We must do both. But I urge us to focus a little less on the short-term fate of our assignments and more on the long-term consequences for students.

Some students use forms to help write their papers and do their homework, and yes, this is something to worry about. We want students to learn, and if they’re sticking an assignment in a box and hitting return, they’re not learning as much.

But this is a small part of a larger problem, which is that students entering college now are more likely to graduate into a world transformed by AI. Templates like ChatGPT are already being integrated into word processing programs and search engines. In 10 years, they’ll be as familiar as AutoComplete is to us now. So telling students “Just say no to AI” will not be a sufficient way to prepare them for 2030. Students will use these models and will need to understand them.

There are certainly some contexts, such as closed-book testing, where it is appropriate to say “don’t use AI,” just as we currently say “don’t look up the answer on the web.” But universities will also need to offer courses and assignments that teach students how to understand and use these tools in appropriate and innovative ways.

What are some other uses for AI language models?

Right now, we approach AI the way we often treat new technology: we try to fit it into a niche that already exists. Large language models are widely understood as writing machines, so we think, “Maybe students will use models to write their course papers.” The forms also seem capable of answering questions, so we think, “Maybe they’ll replace search engines.”

The language model is not a library or a copy from the Internet; It is literally just a form of language. People will be disappointed if they expect the language paradigm itself to provide knowledge.

I think we will find more interesting ways to use this technology. Rather than asking old questions for which there are already answers, a fun way to use one of these templates is often to present new evidence you want to analyze, accurately describing the analysis you want to perform.

I like British programmer Simon Willison’s way of putting this, which is that a language model is a word calculator. The model does not contain comprehensive knowledge. But it is a flexible little machine that can follow verbal instructions, convert text and think out loud – so to speak – in writing.

You wouldn’t ask a calculator to run a physics experiment or bridge geometry on its own – similarly, we probably shouldn’t ask a language model to write important documents on its own. But if we can divide the project into well-defined tasks, the language model may make these tasks easier. The model could, for example, read a pile of emails one by one in order to rate their relevance to the question, and then orient itself to condense the most relevant emails into a summary.

In short, these are not substitutes for human writing or human knowledge. They are flexible language conversion tools. We’ll need to learn how to use it, and it’s also possible that we’ll end up using it for analysis more than we do for writing.